Efficient use of clinical EEG data for deep learning in epilepsy
Catarina da Silva Lourenço1, Marleen C Tjepkema-Cloostermans2, Michel J A M van Putten2
1Department of Clinical Neurophysiology, Institute for Technical Medicine, University of Twente, Technical Medical Centre, Enschede, the Netherlands.
Summary
Dataset augmentation techniques, including temporal shifting and varied electroencephalogram (EEG) montages, significantly improved automated detection of Interictal Epileptiform Discharges (IEDs) using deep learning models.
Area of Science:
- Medical technology
- Artificial intelligence in healthcare
- Neurology
Background:
- Automated detection of Interictal Epileptiform Discharges (IEDs) in electroencephalogram (EEG) recordings is crucial for epilepsy diagnosis.
- Deep learning models show promise for IED detection, but are limited by the scarcity of expert-annotated data.
Purpose of the Study:
- To enhance the performance of deep neural networks for automated IED detection.
- To investigate the impact of dataset augmentation techniques on IED detection accuracy.
Main Methods:
- Utilized EEG data from 50 focal epilepsy patients, 49 generalized epilepsy patients, and 67 controls.
- Applied temporal shifting and diverse EEG montages to augment the dataset, increasing samples with IEDs.
- Trained a VGG C convolutional neural network for IED detection.
Main Results:
- Reduced false positive rates from 2.11 to 0.73 detections per minute at optimal sensitivity and specificity.
- Increased sensitivity from 63% to 96% at 99% specificity.
- Improved model robustness against IED position within epochs and montage variations.
Conclusions:
- Dataset augmentation via temporal shifting and varied EEG montages significantly improves deep neural network performance for IED detection.
- This approach can mitigate the need for extensive expert annotation, potentially revolutionizing EEG analysis.
